Go live
Preflight checks, start serving inference, and monitor your node on the network.
Before you go live
Confirm you have completed:
- Installed Kielo Desktop Node
- Signed in and finished onboarding
- Downloaded at least one model and selected it as active
- Stable network and sufficient disk space for logs/cache
Preflight checks
The desktop app runs preflight before accepting traffic. Typical checks:
- GPU available and not blocked by another process
- Active model loaded and responding to a test prompt
- Valid node registration and auth token
- Reachability / NAT status (affects how jobs are delivered)
- Disk and memory headroom
Fix any failed checks before tapping Go Live.
Start serving
- Open the main dashboard in Kielo Desktop Node
- Confirm the correct model is selected
- Run preflight — all checks green
- Click Go Live — the node registers as available with the routing server
- The app begins heartbeat and job polling against the API gateway
While live
Monitor in the desktop app:
- Active and completed inference jobs
- Token throughput and latency
- GPU temperature and utilization
- Earnings accrual and wallet balance
The web provider dashboard at /provider mirrors earnings, jobs, and performance for when you are away from the desktop app.
Stop serving
Use Stop or go offline from the dashboard to drain jobs gracefully. The node stops accepting new work; in-flight requests complete according to gateway timeout settings.
Network test
After going live, the app may run a network test against the Kielo routing server to validate end-to-end connectivity. Failed tests usually indicate firewall, NAT, or DNS issues — check local network settings and retry.
Session analytics (coming soon)
Future releases will print a shareable session link (e.g. app.kielo.in/a/<hash>) for real-time node analytics during a serving session. Not yet available in v0.1.0.
Troubleshooting
- No jobs routed — verify go-live status, model compatibility, and contribution preferences (availability hours, workload types).
- Auth errors — sign out and sign in again; check token expiry.
- OOM during jobs — switch to a smaller model or higher quantization.